using pretrained model including fc7 layer

I'd like to run some codes below to use trained model to inference test set.
The thing is I saved full model to predict class not embedding. (ckpt_embedding = False)

sym, arg_params, aux_params = mx.model.load_checkpoint(prefix, epoch)
model = mx.mod.Module(symbol=sym, label_names=None, context=mx.cpu())
model.bind(for_training=False, data_shapes=[('data', (1,3,112,112)])
model.set_params(arg_params, aux_params, allow_missing=True)

but I got MXNetError
simple_bind error. Arguments:
data: (1, 3, 112, 112)
MXNetError: Error in operator one_hot0 ... : Index type must be set for one_hot operator

Does anyone know how to get class with pretrained full model?